TL;DR: Emissions verification has become a financial close dependency. Your finance team needs a carbon liability figure on a fixed date, and the verified figure usually arrives later and differs from the one already booked. The fix is not a faster verifier: it is validating fuel data before conversion, running voyage-level verification through the period rather than at the end, and locking the set once it is verified. ZeroNorth's AI-enabled fuel model achieves a median accuracy above 90% across the fleet.
A commercial operations director is asked for one number before the books close: what the year's EU allowance exposure costs. The compliance team has a figure. Finance has a different one, booked months earlier from the voyage system. The verifier has not finished.
None of the three is wrong. They were produced at different points, from data that was still moving. But an auditor reading two figures for the same liability sees an inconsistency, not a timing difference, and inconsistencies become findings.
So the question to put to any emissions platform is not how accurately it calculates. It is when its figures become final, and what happens to them afterwards.
Why the verified figure and the booked figure drift apart
They drift because the carbon number is produced twice, months apart, from a dataset that keeps changing between the two.
The first version is produced during the year, voyage by voyage, so the business can price exposure into commercial decisions and finance can accrue against it. The second is produced at verification, when a third party works through the full period against the regulatory frameworks. Between those two points, late noon reports arrive, bunker delivery notes are reconciled, sensor gaps get backfilled and voyages are reclassified. Every one of those is a legitimate correction. Collectively they move the total.
Four frameworks now sit on the same underlying dataset: EU MRV, UK MRV, IMO DCS and FuelEU Maritime. They have different boundaries, different definitions and different deadlines, which means one dataset has to satisfy four sets of rules without being rebuilt four times. The arithmetic is not the difficulty. Holding one defensible version of the fuel data while four frameworks and a finance team all draw on it is the difficulty.
What has to be true before finance will take the number
The figure has to be validated before it is converted, traceable after it is converted, and frozen once it is verified.
Those three conditions do most of the work, and they are worth testing directly in any evaluation.
Validation before conversion. Fuel data arrives from noon reports, flow meters and bunker delivery notes, each with its own failure modes. Checking each figure against physical limits and against the vessel's own history has to happen before the emission factor is applied, because once a bad reading has been converted and aggregated into a fleet total it is no longer visible as a bad reading. It is simply part of the number. ZeroNorth Emission Analytics runs that validation step first and flags what falls outside expected behaviour rather than averaging it away.
One dataset behind both answers. The exposure figure the business prices against and the figure submitted for verification should be the same figure at different stages of maturity, not two numbers maintained in parallel. Where they are maintained separately, closing the year means a reconciliation exercise that nobody budgeted for, and the reconciliation itself becomes something the auditor examines.
A verified set that stops moving. Once a period has been verified, the underlying data needs to be locked, so that a figure pulled in March matches the figure pulled in June. Without that, there is no stable reference for the accounts, and every subsequent question reopens the whole period.
A useful question when you evaluate: ask what the system does with a late report that arrives after a period has been verified. Whether it is rejected, held in a subsequent period or silently absorbed into the verified total is a design decision, and the three have very different consequences at audit.
Which closing approach you are actually running
Most operators are running one of three patterns, and the one you are on decides what you can put in the accounts.
The third row is the only one that answers the two questions an auditor always asks, which are what the figure was as at the reporting date and whether it can be reproduced now.
Two values for the same fuel, held separately
The calorific value your performance model should use and the calorific value the regulation specifies are often different numbers, and a system that forces you to choose one will make the other wrong.
This surfaces most sharply on gas carriers, where boil-off composition varies by cargo and by terminal, so the real energy content of the fuel burned can sit some way from the default the regulation prescribes. Use the regulatory default everywhere and the performance picture understates what the vessel actually achieved. Use the measured value everywhere and the compliance submission is out of regulation.
The resolution is to hold both and govern them separately: the measured value feeding vessel performance, the prescribed default feeding the EU ETS and FuelEU calculations, from the same reported dataset. The distinction matters commercially because the two figures answer to different audiences. One has to persuade a technical team that the ship is performing; the other has to satisfy a regulator who will not accept a substituted constant, however better founded it is.
What independent verification of the model changes
Verifying the model removes the derivation of the fuel figure from the audit conversation, which is where most of the time goes.
This is the difference between a number you defend and a number you present. An assurance provider examines your boundary, your completeness and your controls, and takes the calculation model as an input. If that model has already been tested by someone outside the commercial relationship, the question of how a modelled consumption figure was arrived at is settled before the exercise starts.
ZeroNorth's AI-enabled fuel model achieves a median accuracy above 90% across the fleet. What model verification does not cover is your own disclosure, which is a separate exercise: we set out the distinction in how emissions figures are calculated and verified.
The bottom line
Judge an emissions platform on when its figures become final and whether they stay final, not on the accuracy it claims for them.
The vendor-neutral question to put to anyone you are evaluating, including us, is this: for a period you have verified, show me the figure as at the reporting date, show me the voyage records behind it, and tell me what happens to a report that arrives tomorrow.
For ZeroNorth the answer is four specifics. Fuel data is validated against physical limits and the vessel's own history before any emission factor is applied. The exposure figure and the submitted figure come from one dataset rather than two that have to be reconciled. Measured and regulatory calorific values are held separately, so performance and compliance can each use the right one. And ZeroNorth's AI-enabled fuel model achieves a median accuracy above 90% across the fleet. Where voyages sit on tonnage you charter rather than own, the evidence problem is different again, and we cover it in what an auditor will accept for voyages you do not control.
If your reporting period is closing now, see how Emission Analytics handles verification and locking, review pricing, or book a demo and bring the period you are closing.
